Fran Rodriguez Ruiz
Thursday 21st March 2019
Time: 4.00pm
Ground Floor Seminar Room
25 Howland Street, London, W1T 4JG
The Probabilistic Modeling Pipeline: Analyzing Consumer Behavior with SHOPPER
Probabilistic models are an effective approach for understanding real-world data. They let us describe our assumptions about the data-generating process using both observed and hidden variables, and they provide the statistical tools we need to unveil the hidden patterns that generated the data. In this talk, I will show how to apply the probabilistic modeling pipeline through an application in econometrics. I will describe SHOPPER, a probabilistic model of shopping data that captures complex econometric assumptions. SHOPPER uses interpretable components to model the latent forces that drive how customers choose products. To discover how those forces are manifest in large-scale consumer data, SHOPPER requires a scalable posterior inference algorithm. I will sketch the main challenges of posterior inference and the advances in variational inference that allow us to apply SHOPPER on data from a major grocery store in the United States. SHOPPER finds meaningful latent representations of items and allows us to answer counterfactual queries about the customers' response to price changes.
Biography
Francisco J. R. Ruiz is a Postdoctoral Research Scientist at the Department of Computer Science in Columbia University and at the Engineering Department in the University of Cambridge. Francisco holds a Marie-Skłodowska Curie Individual Fellowship in the context of the E.U. Horizon 2020 program. He completed his Ph.D. in 2015 and his M.Sc. in 2012, both from the University Carlos III in Madrid. His research is focused on statistical machine learning; in particular, his interests include: approximate Bayesian inference, probabilistic modeling for discrete data, applications of Bayesian non-parametrics, and time series models.